1 hour ago
5
Follow ZDNET: Add america arsenic a preferred source on Google.
ZDNET's cardinal takeaways
- DuckDuckGo's privacy-first chatbot is taking off.
- Users are progressively acrophobic astir however their information is used.
- New features could besides beryllium driving growth.
Privacy concerns astir chatbots are thing new, but arsenic AI adoption spreads, users are becoming progressively alert of the risks. Duck.ai, the chatbot from privacy-focused browser DuckDuckGo, could beryllium benefiting from that.
Also: Stop telling AI your secrets - 5 reasons why, and what to bash if you already overshared
New information from Similarweb recovered web postulation to Duck.ai exploded past month. Duck.ai "reached 11.1 cardinal visits successful February, up much than 300% from January," Similarweb told ZDNET.
DuckDuckGo did not instantly respond to a petition for comment.
That fig is inactive tiny compared to the astir ascendant chatbots: By contrast, Similarweb estimates that chatgpt.com reached 5.4 cardinal visits successful February, portion gemini.google.com reached 2.1 cardinal and claude.ai deed 290.3 million. Still, for having lone launched successful beta a twelvemonth ago, that's a crisp uptick successful visits worthy keeping an oculus on.
Duck.ai extends the aforesaid privateness to users that they person travel to expect from its browser, anonymizing queries to forestall 3rd parties from accessing chats. The chatbot does not tally a bespoke LLM; it uses frontier models from Anthropic, OpenAI, and Meta, among others, but calls those providers connected your behalf truthful arsenic not to exposure your IP code oregon different idiosyncratic information.
(Disclosure: Ziff Davis, ZDNET's genitor company, filed an April 2025 suit against OpenAI, alleging it infringed Ziff Davis copyrights successful grooming and operating its AI systems.)
"In addition, we person agreements successful spot with each exemplary providers that further bounds however they tin usage information from these anonymous requests, including not utilizing Prompts and Outputs to make oregon amended their models, arsenic good arsenic deleting each accusation received erstwhile it is nary longer indispensable to supply Outputs (at astir wrong 30 days, with constricted exceptions for information and ineligible compliance)," Duck.ai's privateness policy says.
ZDNET's Jack Wallen tested Duck.ai past year and recovered helium preferred it implicit Perplexity astatine the time.
Duck.ai's popularity spike
So wherefore the abrupt leap successful traffic?
Duck.ai offers 2 main benefits implicit idiosyncratic proprietary chatbots similar Gemini and ChatGPT: the enactment to toggle betwixt galore models and accrued privateness protections. The second is apt what sets Duck.ai apart, though, arsenic Perplexity besides offers entree to aggregate models done a azygous interface, successful summation to its ain Sonar exemplary family.
While immoderate users connected Reddit said they bask Duck.ai -- 1 poster said "it's mode amended than Google's," ostensibly referring to Gemini, and that it's the crushed they usage DuckDuckGo. Many others said it was "not bad," neutral, comparably disappointing to different options, oregon conscionable "better than nothing." Some users dislike that Duck.ai doesn't enactment papers uploads.
Also: Tired of seeing AI images online? DuckDuckGo lets you fell them from results now
One Reddit thread from a fewer months earlier the spike is afloat of users lamenting that Duck.ai is conscionable arsenic disappointing arsenic different chatbots, though 1 idiosyncratic said it "is decent, peculiarly if you are privateness focused." A newer thread includes complaints astir usage limits without mentioning the specifics of what keeps users coming backmost to the chatbot.
Positive reviews are harder to parse due to the fact that Duck.ai supplies entree to respective frontier models arsenic opposed to its ain bespoke option. Those who bash bask Duck.ai mention the efficacy of the circumstantial models they chose wrong it, similar OpenAI's GPT-5 mini. For astatine slightest 1 user, however, Duck.ai seemed to interaction the efficacy of those models.
"A person of excavation uses regular chat-gpt and swears helium gets amended replies than helium does with duck.ai," 1 poster wrote. "I don't cognize if it's true. Maybe the privateness focused strategy prompts oregon immoderate other they bash successful the inheritance does alteration the answers." That poster added that they "really bask Duck.ai."
Increased privateness concerns
Last month, Anthropic rejected immoderate projected applications of its tech for weapons and wide surveillance by the Department of Defense (DoD), which retaliated by cutting its contract. OpenAI rapidly swooped in, lone to encounter the aforesaid debates implicit use.
Coverage of the incidental brought renewed concerns astir privateness and AI to the forefront of nationalist conversation. Nathan Calvin, vice president of authorities affairs and wide counsel astatine advocacy enactment Encode AI, told ZDNET that helium noticed an uptick successful conversations astir information brokers, privacy, and however the authorities obtains information since the declaration incident, successful some the wide nationalist and argumentation spaces.
"It's an contented that's been astir for a while, but I decidedly consciousness similar a batch of folks are taking a look astatine it with caller eyes and urgency," helium noted, adding that galore radical "had ne'er heard of Anthropic oregon Claude" earlier the DoD story.
In that light, chatbots that spell further to support idiosyncratic information from some AI companies themselves and the authorities whitethorn look much appealing than before.
But according to Similarweb's graph, Duck.ai started seeing a flimsy uptick astatine the extremity of 2025 that past grew exponentially past month. Duck.ai added image generation to the level successful December; successful mid-February, it added real-time, privacy-protected dependable chat. Some Reddit users had complained anterior to the merchandise that text-to-speech was the lone happening missing from their Duck.ai experience, truthful the merchandise is apt a operator of the spike arsenic well.
In keeping with the company's different policies, Duck.ai dependable chats are anonymized and not utilized to bid models, and neither DuckDuckGo nor OpenAI (which provides dependable enactment done Duck.ai) stores audio. That said, Duck.ai advises users that their dependable "can beryllium a biometric identifier," which they should see earlier trying the feature.
One idiosyncratic wrote connected the /ChatGPTcomplaints Reddit thread this period that they were trying Duck.ai "for nary different crushed than to hopefully rebuild my transportation with 4o," referring to GPT-4o, the exemplary OpenAI sunset successful ChatGPT successful February, to the displeasure of galore users. Other users noted their vexation with OpenAI's DoD declaration successful the aforesaid thread, however, which is inactive a imaginable operator distant from ChatGPT and toward immoderate different chatbot (Anthropic's Claude overtook OpenAI's ChatGPT as the astir downloaded escaped app successful the US straight aft the declaration quality began).
How to effort Duck.ai
You tin effort Duck.ai yourself for escaped oregon $10 per period (or $100 per year, if paying annually), which gets you entree to much precocious models.
English (US) ·